펭귄 템플릿을 사용하여 데이터에 대한 TFX 파이프라인 생성

컬렉션을 사용해 정리하기 내 환경설정을 기준으로 콘텐츠를 저장하고 분류하세요.


소개

이 문서는 TFX Python 패키지와 함께 제공되는 펭귄 템플릿 을 사용하여 자신의 데이터 세트에 대한 TensorFlow Extended(TFX) 파이프라인을 만드는 지침을 제공합니다. 생성된 파이프라인은 처음에 Palmer Penguins 데이터세트를 사용하지만 데이터세트의 파이프라인을 변환합니다.

전제 조건

  • 리눅스 / 맥OS
  • 파이썬 3.6-3.8
  • 주피터 노트북

1단계. 미리 정의된 템플릿을 프로젝트 디렉토리에 복사합니다.

이 단계에서는 TFX의 펭귄 템플릿 에서 파일을 복사하여 작업 파이프라인 프로젝트 디렉터리 및 파일을 만듭니다. 이것을 TFX 파이프라인 프로젝트의 스캐폴드로 생각할 수 있습니다.

핍 업데이트

Colab에서 실행 중인 경우 최신 버전의 Pip가 있는지 확인해야 합니다. 물론 로컬 시스템은 별도로 업데이트할 수 있습니다.

import sys
if 'google.colab' in sys.modules:
  !pip install --upgrade pip

필수 패키지 설치

먼저 TFX와 TensorFlow Model Analysis(TFMA)를 설치합니다.

pip install -U tfx tensorflow-model-analysis

TFX의 버전을 확인해보자.

import tensorflow as tf
import tensorflow_model_analysis as tfma
import tfx

print('TF version: {}'.format(tf.__version__))
print('TFMA version: {}'.format(tfma.__version__))
print('TFX version: {}'.format(tfx.__version__))
TF version: 2.7.1
TFMA version: 0.37.0
TFX version: 1.6.0

파이프라인을 만들 준비가 되었습니다.

PROJECT_DIR 을 환경에 적합한 대상으로 설정하십시오. 기본값은 Google Cloud AI Platform Notebook 환경에 적합한 ~/imported/${PIPELINE_NAME} 입니다.

아래 PIPELINE_NAME 을 변경하여 파이프라인에 다른 이름을 지정할 수 있습니다. 이것은 또한 파일이 저장될 프로젝트 디렉토리의 이름이 됩니다.

PIPELINE_NAME="my_pipeline"
import os
# Set this project directory to your new tfx pipeline project.
PROJECT_DIR=os.path.join(os.path.expanduser("~"), "imported", PIPELINE_NAME)

템플릿 파일을 복사합니다.

TFX에는 TFX python 패키지와 함께 penguin 템플릿이 포함되어 있습니다. penguin 템플릿에는 이 튜토리얼의 목적인 파이프라인으로 데이터세트를 가져오기 위한 많은 지침이 포함되어 있습니다.

tfx template copy CLI 명령은 미리 정의된 템플릿 파일을 프로젝트 디렉터리에 복사합니다.

# Set `PATH` to include user python binary directory and a directory containing `skaffold`.
PATH=%env PATH
%env PATH={PATH}:/home/jupyter/.local/bin

!tfx template copy \
  --pipeline-name={PIPELINE_NAME} \
  --destination-path={PROJECT_DIR} \
  --model=penguin
env: PATH=/tmpfs/src/tf_docs_env/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/opt/puppetlabs/bin:/opt/android-studio/current/bin:/usr/local/go/bin:/usr/local/go/packages/bin:/opt/kubernetes/client/bin/:/home/kbuilder/.local/bin:/home/jupyter/.local/bin
CLI
Copying penguin pipeline template
kubeflow_runner.py -> /home/kbuilder/imported/my_pipeline/kubeflow_runner.py
configs.py -> /home/kbuilder/imported/my_pipeline/pipeline/configs.py
pipeline.py -> /home/kbuilder/imported/my_pipeline/pipeline/pipeline.py
__init__.py -> /home/kbuilder/imported/my_pipeline/pipeline/__init__.py
model.py -> /home/kbuilder/imported/my_pipeline/models/model.py
features.py -> /home/kbuilder/imported/my_pipeline/models/features.py
features_test.py -> /home/kbuilder/imported/my_pipeline/models/features_test.py
preprocessing_test.py -> /home/kbuilder/imported/my_pipeline/models/preprocessing_test.py
preprocessing.py -> /home/kbuilder/imported/my_pipeline/models/preprocessing.py
model_test.py -> /home/kbuilder/imported/my_pipeline/models/model_test.py
__init__.py -> /home/kbuilder/imported/my_pipeline/models/__init__.py
constants.py -> /home/kbuilder/imported/my_pipeline/models/constants.py
local_runner.py -> /home/kbuilder/imported/my_pipeline/local_runner.py
__init__.py -> /home/kbuilder/imported/my_pipeline/__init__.py

이 노트북의 작업 디렉토리 컨텍스트를 프로젝트 디렉토리로 변경하십시오.

%cd {PROJECT_DIR}
/home/kbuilder/imported/my_pipeline

복사한 소스 파일 찾아보기

TFX 템플릿은 Python 소스 코드 및 샘플 데이터를 포함하여 파이프라인을 구축하기 위한 기본 스캐폴드 파일을 제공합니다. penguin 템플릿은 펭귄 예제 와 동일한 Palmer Penguins 데이터 세트 및 ML 모델을 사용합니다.

다음은 각 Python 파일에 대한 간략한 소개입니다.

  • pipeline - 이 디렉토리에는 파이프라인의 정의가 포함되어 있습니다.
    • configs.py — 파이프라인 러너에 대한 공통 상수를 정의합니다.
    • pipeline.py .py — TFX 구성 요소 및 파이프라인을 정의합니다.
  • models - 이 디렉터리에는 ML 모델 정의가 포함되어 있습니다.
    • features.py , features_test.py — 모델의 기능을 정의합니다.
    • preprocessing.py , preprocessing_test.py — 데이터에 대한 전처리 루틴을 정의합니다.
    • constants.py — 모델의 상수를 정의합니다.
    • model.py , model_test.py — TensorFlow와 같은 ML 프레임워크를 사용하여 ML 모델을 정의합니다.
  • local_runner.py — 로컬 오케스트레이션 엔진을 사용하는 로컬 환경용 러너 정의
  • kubeflow_runner.py — Kubeflow Pipelines 오케스트레이션 엔진에 대한 러너 정의

기본적으로 템플릿에는 표준 TFX 구성 요소만 포함됩니다. 사용자 지정 작업이 필요한 경우 파이프라인에 대한 사용자 지정 구성 요소를 생성할 수 있습니다. 자세한 내용은 TFX 사용자 지정 구성 요소 가이드 를 참조하세요.

단위 테스트 파일.

이름에 _test.py 가 포함된 일부 파일이 있음을 알 수 있습니다. 이는 파이프라인의 단위 테스트이며 고유한 파이프라인을 구현할 때 더 많은 단위 테스트를 추가하는 것이 좋습니다. -m 플래그와 함께 테스트 파일의 모듈 이름을 제공하여 단위 테스트를 실행할 수 있습니다. 일반적으로 .py 확장자를 삭제하고 / 를 로 대체하여 모듈 이름을 얻을 수 있습니다 . . 예를 들어:

import sys
!{sys.executable} -m models.features_test
Running tests under Python 3.7.5: /tmpfs/src/tf_docs_env/bin/python
[ RUN      ] FeaturesTest.testLabelKey
INFO:tensorflow:time(__main__.FeaturesTest.testLabelKey): 0.0s
I0203 11:08:46.306882 140258321348416 test_util.py:2309] time(__main__.FeaturesTest.testLabelKey): 0.0s
[       OK ] FeaturesTest.testLabelKey
[ RUN      ] FeaturesTest.test_session
[  SKIPPED ] FeaturesTest.test_session
----------------------------------------------------------------------
Ran 2 tests in 0.001s

OK (skipped=1)

로컬 환경에서 TFX 파이프라인을 생성합니다.

TFX는 파이프라인을 실행하기 위해 여러 오케스트레이션 엔진을 지원합니다. 우리는 로컬 오케스트레이션 엔진을 사용할 것입니다. 로컬 오케스트레이션 엔진은 더 이상의 종속성 없이 실행되며, 원격 컴퓨팅 클러스터에 의존하지 않고 로컬 환경에서 실행되기 때문에 개발 및 디버깅에 적합합니다.

local_runner.py 를 사용하여 로컬 오케스트레이터를 사용하여 파이프라인을 실행합니다. 파이프라인을 실행하기 전에 생성해야 합니다. pipeline create 을 생성할 수 있습니다.

tfx pipeline create --engine=local --pipeline_path=local_runner.py
CLI
Creating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" created successfully.

pipeline create 명령은 실제로 실행하지 않고 local_runner.py 에 정의된 파이프라인을 등록합니다.

다음 단계에서 run create 명령으로 생성된 파이프라인을 실행합니다.

2단계. 데이터를 파이프라인에 수집합니다.

초기 파이프라인은 템플릿에 포함된 펭귄 데이터 세트를 수집합니다. 데이터를 파이프라인에 넣어야 하며 대부분의 TFX 파이프라인은 ExampleGen 구성 요소로 시작합니다.

ExampleGen 선택

데이터는 로컬 또는 분산 파일 시스템 또는 쿼리 가능한 시스템에서 파이프라인이 액세스할 수 있는 모든 위치에 저장할 수 있습니다. TFX는 데이터를 TFX 파이프라인으로 가져오는 다양한 ExampleGen 구성 요소 를 제공합니다. 구성 요소를 생성하는 다음 예제에서 하나를 선택할 수 있습니다.

예를 들어 tfx에는 Presto를 데이터 소스로 사용하는 사용자 지정 ExecampleGen이 포함되어 있습니다. 사용자 지정 실행기를 사용하고 개발하는 방법에 대한 자세한 내용 은 가이드 를 참조하세요.

사용할 ExampleGen을 결정했으면 데이터를 사용하도록 파이프라인 정의를 수정해야 합니다.

  1. local_runner.py 에서 DATA_PATH 를 수정하고 파일 위치로 설정합니다.

    • 로컬 환경에 파일이 있는 경우 경로를 지정하십시오. 이것은 파이프라인을 개발하거나 디버깅하기 위한 최상의 옵션입니다.
    • 파일이 GCS에 저장된 경우 gs://{bucket_name}/... 으로 시작하는 경로를 사용할 수 있습니다. 예를 들어 gsutil 을 사용하여 터미널에서 GCS에 액세스할 수 있는지 확인하십시오. 필요한 경우 Google Cloud의 인증 가이드를 따르세요.
    • BigQueryExampleGen과 같은 Query 기반 ExampleGen을 사용하려면 데이터 소스에서 데이터를 선택하기 위한 Query 문이 필요합니다. Google Cloud BigQuery를 데이터 소스로 사용하기 위해 설정해야 할 몇 가지 사항이 더 있습니다.
    • pipeline/configs.py :
      • GOOGLE_CLOUD_PROJECTGCS_BUCKET_NAME 을 GCP 프로젝트 및 버킷 이름으로 변경합니다. 파이프라인을 실행하기 전에 버킷이 있어야 합니다.
      • BIG_QUERY_WITH_DIRECT_RUNNER_BEAM_PIPELINE_ARGS 변수의 주석 처리를 제거합니다.
      • 주석을 제거하고 BIG_QUERY_QUERY 변수를 쿼리 문 으로 설정합니다.
    • local_runner.py 에서 :
      • pipeline.create_pipeline() 에서 대신 data_path 인수를 주석 처리하고 query 인수를 주석 해제합니다.
    • pipeline/pipeline.py 에서:
      • create_pipeline() 에서 data_path 인수를 주석 처리하고 query 인수를 주석 해제합니다.
      • CsvExampleGen 대신 BigQueryExampleGen 을 사용하세요.
  2. 기존 CsvExampleGen을 pipeline/pipeline.py 의 ExampleGen 클래스로 교체합니다. 각 ExampleGen 클래스에는 서로 다른 서명이 있습니다. 자세한 내용은 ExampleGen 구성 요소 가이드 를 참조하세요. pipeline/pipeline.pyimport 문으로 필수 모듈을 가져오는 것을 잊지 마십시오.

초기 파이프라인은 ExampleGen , StatisticsGen , SchemaGenExampleValidator 의 네 가지 구성 요소로 구성됩니다. StatisticsGen , SchemaGenExampleValidator 에 대해 아무것도 변경할 필요가 없습니다. 처음으로 파이프라인을 실행해 보겠습니다.

# Update and run the pipeline.
!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}
CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 1
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=1, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/1"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'output_data_format': 6, 'output_file_format': 5, 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/1/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:09:12.120566', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/1/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:12.120566')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:09:12.848598153    5127 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 1 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/1"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 1
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 2
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=2, input_dict={'examples': [Artifact(artifact: id: 1
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/1"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886553302
last_update_time_since_epoch: 1643886553302
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/2/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:09:12.120566', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/2/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:12.120566')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 2 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 2
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 3
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=3, input_dict={'statistics': [Artifact(artifact: id: 2
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/2"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886556588
last_update_time_since_epoch: 1643886556588
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/3"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'infer_feature_shape': 1, 'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/3/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:09:12.120566', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/3/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:12.120566"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:12.120566"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:12.120566')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/3/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 3 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/3"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:12.120566:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 3
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

"Component ExampleValidator가 완료되었습니다."가 표시되어야 합니다. 파이프라인이 성공적으로 실행된 경우.

파이프라인의 출력을 검사합니다.

TFX 파이프라인은 아티팩트와 파이프라인 실행의 메타데이터를 포함하는 메타데이터 DB(MLMD) 와 아티팩트의 두 가지 출력을 생성합니다. 출력 위치는 local_runner.py 에 정의되어 있습니다. 기본적으로 아티팩트는 tfx_pipeline_output 디렉토리에 저장되고 메타데이터는 tfx_metadata 디렉토리 아래에 sqlite 데이터베이스로 저장됩니다.

MLMD API를 사용하여 이러한 출력을 검사할 수 있습니다. 먼저 방금 생성된 출력 아티팩트를 검색하기 위한 몇 가지 유틸리티 함수를 정의합니다.

import tensorflow as tf
import tfx
from ml_metadata import errors
from ml_metadata.proto import metadata_store_pb2
from tfx.types import artifact_utils

# TODO(b/171447278): Move these functions into TFX library.

def get_latest_executions(store, pipeline_name, component_id = None):
  """Fetch all pipeline runs."""
  if component_id is None:  # Find entire pipeline runs.
    run_contexts = [
        c for c in store.get_contexts_by_type('run')
        if c.properties['pipeline_name'].string_value == pipeline_name
    ]
  else:  # Find specific component runs.
    run_contexts = [
        c for c in store.get_contexts_by_type('component_run')
        if c.properties['pipeline_name'].string_value == pipeline_name and
           c.properties['component_id'].string_value == component_id
    ]
  if not run_contexts:
    return []
  # Pick the latest run context.
  latest_context = max(run_contexts,
                       key=lambda c: c.last_update_time_since_epoch)
  return store.get_executions_by_context(latest_context.id)

def get_latest_artifacts(store, pipeline_name, component_id = None):
  """Fetch all artifacts from latest pipeline execution."""
  executions = get_latest_executions(store, pipeline_name, component_id)

  # Fetch all artifacts produced from the given executions.
  execution_ids = [e.id for e in executions]
  events = store.get_events_by_execution_ids(execution_ids)
  artifact_ids = [
      event.artifact_id for event in events
      if event.type == metadata_store_pb2.Event.OUTPUT
  ]
  return store.get_artifacts_by_id(artifact_ids)

def find_latest_artifacts_by_type(store, artifacts, artifact_type):
  """Get the latest artifacts of a specified type."""
  # Get type information from MLMD
  try:
    artifact_type = store.get_artifact_type(artifact_type)
  except errors.NotFoundError:
    return []
  # Filter artifacts with type.
  filtered_artifacts = [aritfact for aritfact in artifacts
                        if aritfact.type_id == artifact_type.id]
  # Convert MLMD artifact data into TFX Artifact instances.
  return [artifact_utils.deserialize_artifact(artifact_type, artifact)
      for artifact in filtered_artifacts]


from tfx.orchestration.experimental.interactive import visualizations

def visualize_artifacts(artifacts):
  """Visualizes artifacts using standard visualization modules."""
  for artifact in artifacts:
    visualization = visualizations.get_registry().get_visualization(
        artifact.type_name)
    if visualization:
      visualization.display(artifact)

from tfx.orchestration.experimental.interactive import standard_visualizations
standard_visualizations.register_standard_visualizations()

import pprint

from tfx.orchestration import metadata
from tfx.types import artifact_utils
from tfx.types import standard_artifacts

def preview_examples(artifacts):
  """Preview a few records from Examples artifacts."""
  pp = pprint.PrettyPrinter()
  for artifact in artifacts:
    print("==== Examples artifact:{}({})".format(artifact.name, artifact.uri))
    for split in artifact_utils.decode_split_names(artifact.split_names):
      print("==== Reading from split:{}".format(split))
      split_uri = artifact_utils.get_split_uri([artifact], split)

      # Get the list of files in this directory (all compressed TFRecord files)
      tfrecord_filenames = [os.path.join(split_uri, name)
                            for name in os.listdir(split_uri)]
      # Create a `TFRecordDataset` to read these files
      dataset = tf.data.TFRecordDataset(tfrecord_filenames,
                                        compression_type="GZIP")
      # Iterate over the first 2 records and decode them.
      for tfrecord in dataset.take(2):
        serialized_example = tfrecord.numpy()
        example = tf.train.Example()
        example.ParseFromString(serialized_example)
        pp.pprint(example)

import local_runner

metadata_connection_config = metadata.sqlite_metadata_connection_config(
              local_runner.METADATA_PATH)

이제 MLMD에서 출력 아티팩트의 메타데이터를 읽을 수 있습니다.

with metadata.Metadata(metadata_connection_config) as metadata_handler:
    # Search all aritfacts from the previous pipeline run.
    artifacts = get_latest_artifacts(metadata_handler.store, PIPELINE_NAME)
    # Find artifacts of Examples type.
    examples_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.Examples.TYPE_NAME)
    # Find artifacts generated from StatisticsGen.
    stats_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.ExampleStatistics.TYPE_NAME)
    # Find artifacts generated from SchemaGen.
    schema_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.Schema.TYPE_NAME)
    # Find artifacts generated from ExampleValidator.
    anomalies_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.ExampleAnomalies.TYPE_NAME)

이제 각 구성 요소의 출력을 검사할 수 있습니다. TFDV(Tensorflow Data Validation)StatisticsGen , SchemaGenExampleValidator 에서 사용되며 TFDV는 이러한 구성 요소의 출력을 시각화하는 데 사용할 수 있습니다.

이 자습서에서는 TFDV를 내부적으로 사용하여 시각화를 표시하는 TFX의 시각화 도우미 메서드를 사용합니다. 각 구성 요소에 대한 자세한 내용은 TFX 구성 요소 자습서 를 참조하십시오.

출력 양식 보기 ExampleGen

ExampleGen의 출력을 살펴보겠습니다. 각 분할에 대한 처음 두 가지 예를 살펴보십시오.

preview_examples(examples_artifacts)

기본적으로 TFX ExampleGen은 예제를 traineval 의 두 분할로 나누지만 분할 구성을 조정할 수 있습니다 .

StatisticsGen의 출력 검사

visualize_artifacts(stats_artifacts)

이러한 통계는 SchemaGen에 제공되어 데이터 스키마를 자동으로 구성합니다.

SchemaGen의 출력 검사

visualize_artifacts(schema_artifacts)

이 스키마는 StatisticsGen의 출력에서 ​​자동으로 유추됩니다. 이 자습서에서는 생성된 스키마를 사용하지만 스키마를 수정하고 사용자 지정할 수도 있습니다.

ExampleValidator의 출력 검사

visualize_artifacts(anomalies_artifacts)

이상이 발견되면 모든 예가 가정을 따른다는 데이터를 검토할 수 있습니다. StatistcsGen과 같은 다른 구성 요소의 출력이 유용할 수 있습니다. 발견된 이상은 파이프라인 실행을 차단하지 않습니다.

SchemaGen 의 출력에서 ​​사용 가능한 기능을 볼 수 있습니다. 기능을 사용하여 Trainer 에서 직접 ML 모델을 구성할 수 있는 경우 다음 단계를 건너뛰고 4단계로 이동할 수 있습니다. 그렇지 않으면 다음 단계에서 일부 기능 엔지니어링 작업을 수행할 수 있습니다. Transform 구성 요소는 평균 계산과 같은 전체 전달 작업이 필요할 때, 특히 확장해야 할 때 필요합니다.

3단계. (선택 사항) Transform 구성 요소를 사용한 기능 엔지니어링.

이 단계에서는 파이프라인의 Transform 구성 요소에서 사용할 다양한 기능 엔지니어링 작업을 정의합니다. 자세한 내용은 Transform 구성 요소 가이드 를 참조하세요.

이는 ExampleGen의 출력에서 ​​사용할 수 없는 추가 기능이 훈련 코드에 필요한 경우에만 필요합니다. 그렇지 않으면 Trainer를 사용하는 다음 단계로 빠르게 이동하십시오.

모델의 기능 정의

models/features.py 에는 기능 이름, 어휘 크기 등을 포함하여 모델의 기능을 정의하는 상수가 포함되어 있습니다. 기본적으로 penguin 템플릿에는 FEATURE_KEYSLABEL_KEY 라는 두 개의 비용이 있습니다. 왜냐하면 penguin 모델은 지도 학습을 사용하여 분류 문제를 해결하고 모든 기능이 연속 숫자 기능이기 때문입니다. 다른 예 는 시카고 택시 예의 기능 정의를 참조하십시오.

preprocessing_fn()에서 훈련/서빙을 위한 전처리를 구현합니다.

실제 기능 엔지니어링은 models/preprocessing.pypreprocessing_fn() 함수에서 발생합니다.

preprocessing_fn 에서 텐서의 입력 딕셔너리를 조작하여 텐서의 출력 딕셔너리를 생성하는 일련의 함수를 정의할 수 있습니다. TensorFlow Transform API에는 scale_to_0_1compute_and_apply_vocabulary 와 같은 도우미 함수가 있거나 단순히 일반 TensorFlow 함수를 사용할 수 있습니다. 기본적으로 penguin 템플릿에는 기능 값을 정규화하기 위한 tft.scale_to_z_score 함수의 사용 예가 포함되어 있습니다.

preprocessing_fn 작성에 대한 자세한 내용은 Tensflow Transform 가이드 를 참조하세요.

파이프라인에 변환 구성 요소를 추가합니다.

preprocessing_fn이 준비되면 파이프라인에 Transform 구성 요소를 추가합니다.

  1. pipeline/pipeline.py 파일에서 # components.append(transform) 주석을 제거하여 파이프라인에 구성 요소를 추가합니다.

파이프라인을 업데이트하고 다시 실행할 수 있습니다.

!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}
CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 4
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=4, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/4"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'output_file_format': 5, 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_data_format': 6, 'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/4/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:09:37.055994', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/4/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:37.055994')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:09:37.596944686    5287 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 4 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/4"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 4
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 5
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=5, input_dict={'examples': [Artifact(artifact: id: 4
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/4"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886578210
last_update_time_since_epoch: 1643886578210
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/5/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:09:37.055994', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/5/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:37.055994')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 5 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 5
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 6
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=6, input_dict={'statistics': [Artifact(artifact: id: 5
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/5"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886581527
last_update_time_since_epoch: 1643886581527
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/6"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'exclude_splits': '[]', 'infer_feature_shape': 1}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/6/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:09:37.055994', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/6/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:09:37.055994"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:09:37.055994"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:09:37.055994')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/6/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 6 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/6"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:09:37.055994:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 6
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

파이프라인이 성공적으로 실행되면 "구성 요소 변환이 완료되었습니다."가 표시되어야 합니다. 로그 어딘가에 . Transform 구성 요소와 ExampleValidator 구성 요소는 서로 종속되지 않으므로 실행 순서가 고정되어 있지 않습니다. 즉, TransformExampleValidator 중 하나가 파이프라인 실행의 마지막 구성 요소가 될 수 있습니다.

변환의 출력 검사

Transform 구성 요소는 Tensorflow 그래프와 변환된 예제의 두 가지 출력을 생성합니다. 변환된 예제는 ExampleGen에서도 생성된 예제 아티팩트 유형이지만 여기에는 변환된 특성 값이 대신 포함됩니다.

이전 단계에서 했던 것처럼 이를 검사할 수 있습니다.

with metadata.Metadata(metadata_connection_config) as metadata_handler:
    # Search all aritfacts from the previous run of Transform component.
    artifacts = get_latest_artifacts(metadata_handler.store,
                                     PIPELINE_NAME, "Transform")
    # Find artifacts of Examples type.
    transformed_examples_artifacts = find_latest_artifacts_by_type(
        metadata_handler.store, artifacts,
        standard_artifacts.Examples.TYPE_NAME)
preview_examples(transformed_examples_artifacts)

4단계. Trainer 구성 요소로 모델을 훈련시킵니다.

Trainer 구성 요소를 사용하여 ML 모델을 빌드합니다. 자세한 내용은 트레이너 구성 요소 가이드 를 참조하세요. Trainer 구성 요소에 모델 코드를 제공해야 합니다.

모델을 정의합니다.

펭귄 템플릿에서 models.model.run_fnTrainer 컴포넌트의 run_fn 인수로 사용됩니다. Trainer 컴포넌트가 실행될 때 models/model.pyrun_fn() 함수가 호출된다는 의미입니다. 주어진 코드에서 keras API를 사용하여 간단한 DNN 모델을 구성하는 코드를 볼 수 있습니다. TFX에서 keras API를 사용하는 방법에 대한 자세한 내용은 TFX 가이드의 TensorFlow 2.x를 참조하세요.

run_fn 에서 모델을 빌드하고 구성 요소에서 지정하는 fn_args.serving_model_dir 이 가리키는 디렉터리에 저장해야 합니다. fn_args run_fn 다른 인수를 사용할 수 있습니다. fn_args 의 전체 인수 목록은 관련 코드 를 참조하십시오.

models/features.py 에서 기능을 정의하고 필요에 따라 사용하십시오. 3단계에서 기능을 변환한 경우 변환된 기능을 모델에 대한 입력으로 사용해야 합니다.

파이프라인에 Trainer 구성 요소를 추가합니다.

run_fn이 준비되면 Trainer 구성 요소를 파이프라인에 추가합니다.

  1. pipeline/pipeline.py 파일에서 # components.append(trainer) 주석을 제거하여 파이프라인에 구성 요소를 추가합니다.

트레이너 구성 요소에 대한 인수는 변환 구성 요소를 사용하는지 여부에 따라 다를 수 있습니다.

  • Transform 구성 요소를 사용하지 않는 경우 인수를 변경할 필요가 없습니다.
  • Transform 컴포넌트를 사용하는 경우 Trainer 컴포넌트 인스턴스를 생성할 때 인수를 변경해야 합니다.

    • examples 인수를 examples=transform.outputs['transformed_examples'], 변경합니다. 훈련을 위해 변환된 예제를 사용해야 합니다.
    • transform_graph=transform.outputs['transform_graph'], 와 같은 transform_graph 인수를 추가합니다. 이 그래프에는 변환 작업에 대한 TensorFlow 그래프가 포함되어 있습니다.
    • 위와 같이 변경하면 Trainer 컴포넌트 생성을 위한 코드는 다음과 같을 것입니다.
    # If you use a Transform component.
    trainer = Trainer(
        run_fn=run_fn,
        examples=transform.outputs['transformed_examples'],
        transform_graph=transform.outputs['transform_graph'],
        schema=schema_gen.outputs['schema'],
        ...
    

파이프라인을 업데이트하고 다시 실행할 수 있습니다.

!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}
CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 7
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=7, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/7"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'input_base': '/home/kbuilder/imported/my_pipeline/data', 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_data_format': 6, 'output_file_format': 5, 'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/7/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:10:00.469382', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/7/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:00.469382')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:10:01.173700221    5436 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 7 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/7"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 7
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 8
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=8, input_dict={'examples': [Artifact(artifact: id: 7
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/7"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886601629
last_update_time_since_epoch: 1643886601629
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/8/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:10:00.469382', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/8/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:00.469382')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 8 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 8
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 9
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=9, input_dict={'statistics': [Artifact(artifact: id: 8
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/8"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886605023
last_update_time_since_epoch: 1643886605023
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/9"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'exclude_splits': '[]', 'infer_feature_shape': 1}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/9/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:10:00.469382', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/9/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:00.469382"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:00.469382"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:00.469382')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/9/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 9 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/9"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:00.469382:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 9
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

이 실행이 성공적으로 실행되면 이제 모델에 대한 첫 번째 TFX 파이프라인을 만들고 실행한 것입니다. 축하합니다!

새 모델은 출력 디렉토리 아래의 특정 위치에 위치하지만 많은 중간 결과를 보유하는 TFX 파이프라인 외부의 서비스 또는 고정된 위치에 모델을 두는 것이 좋습니다. ML 프로덕션 시스템에서 중요한 빌드된 모델을 지속적으로 평가하면 더욱 좋습니다. 다음 단계에서 TFX에서 지속적인 평가 및 배포가 작동하는 방식을 살펴보겠습니다.

5단계. (선택 사항) Evaluator로 모델을 평가하고 pusher로 게시합니다.

Evaluator 구성 요소는 Trainer 에서 빌드된 모든 모델을 지속적으로 평가하고 Pusher 는 모델을 파일 시스템의 사전 정의된 위치 또는 Google Cloud AI Platform Models 에 복사합니다.

파이프라인에 Evaluator 구성 요소를 추가합니다.

pipeline/pipeline.py 파일에서:

  1. 파이프라인에 최신 모델 해석기를 추가하려면 # components.append(model_resolver) 주석을 제거하십시오. Evaluator는 모델을 마지막 파이프라인 실행에서 Evaluator를 통과한 이전 기준 모델과 비교하는 데 사용할 수 있습니다. LatestBlessedModelResolver 는 Evaluator를 통과한 최신 모델을 찾습니다.
  2. 모델에 적절한 tfma.MetricsSpec 을 설정하십시오. 평가는 ML 모델마다 다를 수 있습니다. 펭귄 템플릿에서는 다중 범주 분류 문제를 해결하기 때문에 SparseCategoricalAccuracy 가 사용되었습니다. 또한 특정 슬라이스에 대한 모델을 분석하려면 tfma.SliceSpec 을 지정해야 합니다. 자세한 내용은 Evaluator 구성 요소 가이드 를 참조하세요.
  3. 구성 요소를 파이프라인에 추가하려면 # components.append(evaluator) 주석을 제거하십시오.

파이프라인을 업데이트하고 다시 실행할 수 있습니다.

# Update and run the pipeline.
!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}
CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 10
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=10, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/10"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'output_file_format': 5, 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_data_format': 6, 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/10/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:10:24.358660', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/10/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:24.358660')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:10:24.894390124    5584 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 10 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/10"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 10
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 11
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=11, input_dict={'examples': [Artifact(artifact: id: 10
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/10"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886625515
last_update_time_since_epoch: 1643886625515
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/11/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:10:24.358660', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/11/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:24.358660')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 11 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 11
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 12
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=12, input_dict={'statistics': [Artifact(artifact: id: 11
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/11"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886628941
last_update_time_since_epoch: 1643886628941
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/12"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'infer_feature_shape': 1, 'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/12/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:10:24.358660', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/12/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:24.358660"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:24.358660"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:24.358660')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/12/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 12 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/12"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:24.358660:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 12
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

Evaluator의 출력 검토

이 단계에는 TensorFlow Model Analysis(TFMA) Jupyter 노트북 확장이 필요합니다. TFMA 노트북 확장 버전은 TFMA python 패키지 버전과 동일해야 합니다.

다음 명령은 NPM 레지스트리에서 TFMA 노트북 확장을 설치합니다. 완료하는 데 몇 분 정도 걸릴 수 있습니다.

# Install TFMA notebook extension.
jupyter labextension install tensorflow_model_analysis@{tfma.__version__}
usage: jupyter [-h] [--version] [--config-dir] [--data-dir] [--runtime-dir]
               [--paths] [--json] [--debug]
               [subcommand]

Jupyter: Interactive Computing

positional arguments:
  subcommand     the subcommand to launch

optional arguments:
  -h, --help     show this help message and exit
  --version      show the versions of core jupyter packages and exit
  --config-dir   show Jupyter config dir
  --data-dir     show Jupyter data dir
  --runtime-dir  show Jupyter runtime dir
  --paths        show all Jupyter paths. Add --json for machine-readable
                 format.
  --json         output paths as machine-readable json
  --debug        output debug information about paths

Available subcommands: bundlerextension console dejavu execute kernel
kernelspec migrate nbconvert nbextension notebook qtconsole run
serverextension troubleshoot trust

Jupyter command `jupyter-labextension` not found.

설치가 완료되면 브라우저를 새로고침하여 확장 프로그램이 적용되도록 하십시오.

with metadata.Metadata(metadata_connection_config) as metadata_handler:
  # Search all aritfacts from the previous pipeline run.
  artifacts = get_latest_artifacts(metadata_handler.store, PIPELINE_NAME)
  model_evaluation_artifacts = find_latest_artifacts_by_type(
      metadata_handler.store, artifacts,
      standard_artifacts.ModelEvaluation.TYPE_NAME)
if model_evaluation_artifacts:
  tfma_result = tfma.load_eval_result(model_evaluation_artifacts[0].uri)
  tfma.view.render_slicing_metrics(tfma_result)

파이프라인에 푸셔 구성 요소를 추가합니다.

모델이 유망해 보이면 모델을 게시해야 합니다. 푸셔 구성요소커스텀 실행기 를 사용하여 파일 시스템의 위치 또는 GCP AI Platform 모델에 모델을 게시할 수 있습니다.

Evaluator 구성 요소는 Trainer 에서 빌드된 모든 모델을 지속적으로 평가하고 Pusher 는 모델을 파일 시스템의 사전 정의된 위치 또는 Google Cloud AI Platform Models 에 복사합니다.

  1. local_runner.py 에서 SERVING_MODEL_DIR 을 게시할 디렉터리로 설정합니다.
  2. pipeline/pipeline.py 파일에서 # components.append(pusher) 주석을 제거하여 파이프라인에 푸셔를 추가합니다.

파이프라인을 업데이트하고 다시 실행할 수 있습니다.

# Update and run the pipeline.
!tfx pipeline update --engine=local --pipeline_path=local_runner.py \
 && tfx run create --engine=local --pipeline_name={PIPELINE_NAME}
CLI
Updating pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
Pipeline "my_pipeline" updated successfully.
CLI
Creating a run for pipeline: my_pipeline
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Excluding no splits because exclude_splits is not set.
INFO:absl:Using deployment config:
 executor_specs {
  key: "CsvExampleGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.example_gen.csv_example_gen.executor.Executor"
      }
    }
  }
}
executor_specs {
  key: "SchemaGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.schema_gen.executor.Executor"
    }
  }
}
executor_specs {
  key: "StatisticsGen"
  value {
    beam_executable_spec {
      python_executor_spec {
        class_path: "tfx.components.statistics_gen.executor.Executor"
      }
    }
  }
}
custom_driver_specs {
  key: "CsvExampleGen"
  value {
    python_class_executable_spec {
      class_path: "tfx.components.example_gen.driver.FileBasedDriver"
    }
  }
}
metadata_connection_config {
  database_connection_config {
    sqlite {
      filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
      connection_mode: READWRITE_OPENCREATE
    }
  }
}

INFO:absl:Using connection config:
 sqlite {
  filename_uri: "./tfx_metadata/my_pipeline/metadata.db"
  connection_mode: READWRITE_OPENCREATE
}

INFO:absl:Component CsvExampleGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:select span and version = (0, None)
INFO:absl:latest span and version = (0, None)
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 13
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=13, input_dict={}, output_dict=defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/13"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}), exec_properties={'input_config': '{\n  "splits": [\n    {\n      "name": "single_split",\n      "pattern": "*"\n    }\n  ]\n}', 'output_config': '{\n  "split_config": {\n    "splits": [\n      {\n        "hash_buckets": 2,\n        "name": "train"\n      },\n      {\n        "hash_buckets": 1,\n        "name": "eval"\n      }\n    ]\n  }\n}', 'output_file_format': 5, 'output_data_format': 6, 'input_base': '/home/kbuilder/imported/my_pipeline/data', 'span': 0, 'version': None, 'input_fingerprint': 'split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/13/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/stateful_working_dir/2022-02-03T11:10:48.556314', tmp_dir='./tfx_pipeline_output/my_pipeline/CsvExampleGen/.system/executor_execution/13/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.example_gen.csv_example_gen.component.CsvExampleGen"
  }
  id: "CsvExampleGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.CsvExampleGen"
      }
    }
  }
}
outputs {
  outputs {
    key: "examples"
    value {
      artifact_spec {
        type {
          name: "Examples"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          properties {
            key: "version"
            value: INT
          }
          base_type: DATASET
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "input_base"
    value {
      field_value {
        string_value: "/home/kbuilder/imported/my_pipeline/data"
      }
    }
  }
  parameters {
    key: "input_config"
    value {
      field_value {
        string_value: "{\n  \"splits\": [\n    {\n      \"name\": \"single_split\",\n      \"pattern\": \"*\"\n    }\n  ]\n}"
      }
    }
  }
  parameters {
    key: "output_config"
    value {
      field_value {
        string_value: "{\n  \"split_config\": {\n    \"splits\": [\n      {\n        \"hash_buckets\": 2,\n        \"name\": \"train\"\n      },\n      {\n        \"hash_buckets\": 1,\n        \"name\": \"eval\"\n      }\n    ]\n  }\n}"
      }
    }
  }
  parameters {
    key: "output_data_format"
    value {
      field_value {
        int_value: 6
      }
    }
  }
  parameters {
    key: "output_file_format"
    value {
      field_value {
        int_value: 5
      }
    }
  }
}
downstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:48.556314')
INFO:absl:Generating examples.
INFO:absl:Processing input csv data /home/kbuilder/imported/my_pipeline/data/* to TFExample.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
E0203 11:10:49.163841363    5734 fork_posix.cc:70]           Fork support is only compatible with the epoll1 and poll polling strategies
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.
INFO:absl:Examples generated.
INFO:absl:Value type <class 'NoneType'> of key version in exec_properties is not supported, going to drop it
INFO:absl:Value type <class 'list'> of key _beam_pipeline_args in exec_properties is not supported, going to drop it
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 13 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'examples': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/13"
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}) for execution 13
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component CsvExampleGen is finished.
INFO:absl:Component StatisticsGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 14
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=14, input_dict={'examples': [Artifact(artifact: id: 13
type_id: 15
uri: "./tfx_pipeline_output/my_pipeline/CsvExampleGen/examples/13"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "file_format"
  value {
    string_value: "tfrecords_gzip"
  }
}
custom_properties {
  key: "input_fingerprint"
  value {
    string_value: "split:single_split,num_files:1,total_bytes:25648,xor_checksum:1643886522,sum_checksum:1643886522"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:CsvExampleGen:examples:0"
  }
}
custom_properties {
  key: "payload_format"
  value {
    string_value: "FORMAT_TF_EXAMPLE"
  }
}
custom_properties {
  key: "span"
  value {
    int_value: 0
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886649739
last_update_time_since_epoch: 1643886649739
, artifact_type: id: 15
name: "Examples"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
properties {
  key: "version"
  value: INT
}
base_type: DATASET
)]}, output_dict=defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:StatisticsGen:statistics:0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}), exec_properties={'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/14/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/stateful_working_dir/2022-02-03T11:10:48.556314', tmp_dir='./tfx_pipeline_output/my_pipeline/StatisticsGen/.system/executor_execution/14/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.statistics_gen.component.StatisticsGen"
    base_type: PROCESS
  }
  id: "StatisticsGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.StatisticsGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "examples"
    value {
      channels {
        producer_node_query {
          id: "CsvExampleGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.CsvExampleGen"
            }
          }
        }
        artifact_query {
          type {
            name: "Examples"
            base_type: DATASET
          }
        }
        output_key: "examples"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "statistics"
    value {
      artifact_spec {
        type {
          name: "ExampleStatistics"
          properties {
            key: "span"
            value: INT
          }
          properties {
            key: "split_names"
            value: STRING
          }
          base_type: STATISTICS
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
}
upstream_nodes: "CsvExampleGen"
downstream_nodes: "SchemaGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:48.556314')
INFO:absl:Generating statistics for split train.
INFO:absl:Statistics for split train written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14/Split-train.
INFO:absl:Generating statistics for split eval.
INFO:absl:Statistics for split eval written to ./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14/Split-eval.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.7 interpreter.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 14 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'statistics': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}) for execution 14
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component StatisticsGen is finished.
INFO:absl:Component SchemaGen is running.
INFO:absl:Running launcher for node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}

INFO:absl:MetadataStore with DB connection initialized
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Going to run a new execution 15
INFO:absl:Going to run a new execution: ExecutionInfo(execution_id=15, input_dict={'statistics': [Artifact(artifact: id: 14
type_id: 17
uri: "./tfx_pipeline_output/my_pipeline/StatisticsGen/statistics/14"
properties {
  key: "split_names"
  value {
    string_value: "[\"train\", \"eval\"]"
  }
}
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:StatisticsGen:statistics:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
state: LIVE
create_time_since_epoch: 1643886653128
last_update_time_since_epoch: 1643886653128
, artifact_type: id: 17
name: "ExampleStatistics"
properties {
  key: "span"
  value: INT
}
properties {
  key: "split_names"
  value: STRING
}
base_type: STATISTICS
)]}, output_dict=defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/15"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:SchemaGen:schema:0"
  }
}
, artifact_type: name: "Schema"
)]}), exec_properties={'infer_feature_shape': 1, 'exclude_splits': '[]'}, execution_output_uri='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/15/executor_output.pb', stateful_working_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/stateful_working_dir/2022-02-03T11:10:48.556314', tmp_dir='./tfx_pipeline_output/my_pipeline/SchemaGen/.system/executor_execution/15/.temp/', pipeline_node=node_info {
  type {
    name: "tfx.components.schema_gen.component.SchemaGen"
    base_type: PROCESS
  }
  id: "SchemaGen"
}
contexts {
  contexts {
    type {
      name: "pipeline"
    }
    name {
      field_value {
        string_value: "my_pipeline"
      }
    }
  }
  contexts {
    type {
      name: "pipeline_run"
    }
    name {
      field_value {
        string_value: "2022-02-03T11:10:48.556314"
      }
    }
  }
  contexts {
    type {
      name: "node"
    }
    name {
      field_value {
        string_value: "my_pipeline.SchemaGen"
      }
    }
  }
}
inputs {
  inputs {
    key: "statistics"
    value {
      channels {
        producer_node_query {
          id: "StatisticsGen"
        }
        context_queries {
          type {
            name: "pipeline"
          }
          name {
            field_value {
              string_value: "my_pipeline"
            }
          }
        }
        context_queries {
          type {
            name: "pipeline_run"
          }
          name {
            field_value {
              string_value: "2022-02-03T11:10:48.556314"
            }
          }
        }
        context_queries {
          type {
            name: "node"
          }
          name {
            field_value {
              string_value: "my_pipeline.StatisticsGen"
            }
          }
        }
        artifact_query {
          type {
            name: "ExampleStatistics"
            base_type: STATISTICS
          }
        }
        output_key: "statistics"
      }
      min_count: 1
    }
  }
}
outputs {
  outputs {
    key: "schema"
    value {
      artifact_spec {
        type {
          name: "Schema"
        }
      }
    }
  }
}
parameters {
  parameters {
    key: "exclude_splits"
    value {
      field_value {
        string_value: "[]"
      }
    }
  }
  parameters {
    key: "infer_feature_shape"
    value {
      field_value {
        int_value: 1
      }
    }
  }
}
upstream_nodes: "StatisticsGen"
execution_options {
  caching_options {
  }
}
, pipeline_info=id: "my_pipeline"
, pipeline_run_id='2022-02-03T11:10:48.556314')
INFO:absl:Processing schema from statistics for split train.
INFO:absl:Processing schema from statistics for split eval.
INFO:absl:Schema written to ./tfx_pipeline_output/my_pipeline/SchemaGen/schema/15/schema.pbtxt.
INFO:absl:Cleaning up stateless execution info.
INFO:absl:Execution 15 succeeded.
INFO:absl:Cleaning up stateful execution info.
INFO:absl:Publishing output artifacts defaultdict(<class 'list'>, {'schema': [Artifact(artifact: uri: "./tfx_pipeline_output/my_pipeline/SchemaGen/schema/15"
custom_properties {
  key: "name"
  value {
    string_value: "my_pipeline:2022-02-03T11:10:48.556314:SchemaGen:schema:0"
  }
}
custom_properties {
  key: "tfx_version"
  value {
    string_value: "1.6.0"
  }
}
, artifact_type: name: "Schema"
)]}) for execution 15
INFO:absl:MetadataStore with DB connection initialized
INFO:absl:Component SchemaGen is finished.

SERVING_MODEL_DIR 에서 새 모델을 찾을 수 있습니다.

6단계. (선택 사항) 파이프라인을 GCP의 Kubeflow Pipelines에 배포합니다.

앞서 언급했듯이 local_runner.py 는 디버깅 또는 개발 목적으로 적합하지만 프로덕션 워크로드를 위한 최상의 솔루션은 아닙니다. 이 단계에서는 파이프라인을 Google Cloud의 Kubeflow Pipelines에 배포합니다.

준비

Kubeflow Pipelines 클러스터에 파이프라인을 배포하려면 kfp python 패키지와 skaffold 프로그램이 필요합니다.

pip install --upgrade -q kfp

# Download skaffold and set it executable.
curl -Lo skaffold https://storage.googleapis.com/skaffold/releases/latest/skaffold-linux-amd64 && chmod +x skaffold

skaffold 바이너리를 쉘이 찾을 수 있는 곳으로 옮겨야 합니다. 또는 --skaffold-cmd 플래그를 사용하여 tfx 바이너리를 실행할 때 skaffold의 경로를 지정할 수 있습니다.

# Move skaffold binary into your path
mv skaffold /home/jupyter/.local/bin/
mv: cannot move 'skaffold' to '/home/jupyter/.local/bin/': No such file or directory

파이프라인을 실행하려면 Kubeflow Pipelines 클러스터도 필요합니다. TFX on Cloud AI Platform 파이프라인 가이드 의 1단계와 2단계를 따르세요.

클러스터가 준비되면 Google 클라우드 콘솔의 Pipelines 라인 대시보드 열기 를 클릭하여 파이프라인 대시보드를 엽니다. 이 페이지의 URL은 파이프라인 실행을 요청하는 ENDPOINT 입니다. 끝점 값은 https:// 이후의 URL에서 googleusercontent.com까지의 모든 값입니다. 엔드포인트를 다음 코드 블록에 넣습니다.

ENDPOINT='' # Enter your ENDPOINT here.

Kubeflow Pipelines 클러스터에서 코드를 실행하려면 코드를 컨테이너 이미지로 압축해야 합니다. 이미지는 파이프라인을 배포하는 동안 자동으로 빌드되며 이미지의 이름과 컨테이너 레지스트리만 설정하면 됩니다. 이 예에서는 Google Container Registry 를 사용하고 이름을 tfx-pipeline 지정합니다.

# Read GCP project id from env.
shell_output=!gcloud config list --format 'value(core.project)' 2>/dev/null
GOOGLE_CLOUD_PROJECT=shell_output[0]

# Docker image name for the pipeline image.
CUSTOM_TFX_IMAGE='gcr.io/' + GOOGLE_CLOUD_PROJECT + '/tfx-pipeline'

데이터 위치를 설정합니다.

Kubeflow Pipelines 클러스터에서 데이터에 액세스할 수 있어야 합니다. 로컬 환경에서 데이터를 사용한 경우 Google Cloud Storage와 같은 원격 저장소에 데이터를 업로드해야 할 수 있습니다. 예를 들어 다음과 같이 Kubeflow Pipelines 클러스터가 배포될 때 자동으로 생성되는 기본 버킷에 펭귄 데이터를 업로드할 수 있습니다.

gsutil cp data/data.csv gs://{GOOGLE_CLOUD_PROJECT}-kubeflowpipelines-default/tfx-template/data/penguin/
Copying file://data/data.csv [Content-Type=text/csv]...
NotFoundException: 404 The destination bucket gs://tf-benchmark-dashboard-kubeflowpipelines-default does not exist or the write to the destination must be restarted

kubeflow_runner.pyDATA_PATH 에 저장된 데이터 위치를 업데이트합니다.

BigQueryExampleGen을 사용하는 경우 데이터 파일을 업로드할 필요는 없지만 beam_pipeline_argspipeline.create_pipeline() 함수에 대해 동일한 querykubeflow_runner.py 인수를 사용하는지 확인하십시오.

파이프라인을 배포합니다.

모든 것이 준비되면 tfx pipeline create 명령을 사용하여 파이프라인을 생성할 수 있습니다.

!tfx pipeline create  \
--engine=kubeflow \
--pipeline-path=kubeflow_runner.py \
--endpoint={ENDPOINT} \
--build-target-image={CUSTOM_TFX_IMAGE}
CLI
[Error] --build-target-image flag was DELETED. You should specify the build target image at the `KubeflowDagRunnerConfig` class instead, and use --build-image flag without argument to build a container image when creating or updating a pipeline.

이제 tfx run create 명령을 사용하여 새로 생성된 파이프라인으로 실행 실행을 시작합니다.

tfx run create --engine=kubeflow --pipeline-name={PIPELINE_NAME} --endpoint={ENDPOINT}
CLI
Creating a run for pipeline: my_pipeline
Failed to load kube config.
Traceback (most recent call last):
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 175, in _new_conn
    (self._dns_host, self.port), self.timeout, **extra_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/util/connection.py", line 95, in create_connection
    raise err
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/util/connection.py", line 85, in create_connection
    sock.connect(sa)
ConnectionRefusedError: [Errno 111] Connection refused

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 710, in urlopen
    chunked=chunked,
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 398, in _make_request
    conn.request(method, url, **httplib_request_kw)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 239, in request
    super(HTTPConnection, self).request(method, url, body=body, headers=headers)
  File "/usr/lib/python3.7/http/client.py", line 1256, in request
    self._send_request(method, url, body, headers, encode_chunked)
  File "/usr/lib/python3.7/http/client.py", line 1302, in _send_request
    self.endheaders(body, encode_chunked=encode_chunked)
  File "/usr/lib/python3.7/http/client.py", line 1251, in endheaders
    self._send_output(message_body, encode_chunked=encode_chunked)
  File "/usr/lib/python3.7/http/client.py", line 1030, in _send_output
    self.send(msg)
  File "/usr/lib/python3.7/http/client.py", line 970, in send
    self.connect()
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 205, in connect
    conn = self._new_conn()
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connection.py", line 187, in _new_conn
    self, "Failed to establish a new connection: %s" % e
urllib3.exceptions.NewConnectionError: <urllib3.connection.HTTPConnection object at 0x7ff729e34190>: Failed to establish a new connection: [Errno 111] Connection refused

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/tmpfs/src/tf_docs_env/bin/tfx", line 8, in <module>
    sys.exit(cli_group())
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 829, in __call__
    return self.main(*args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 782, in main
    rv = self.invoke(ctx)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 1259, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 1259, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 1066, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 610, in invoke
    return callback(*args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/decorators.py", line 73, in new_func
    return ctx.invoke(f, obj, *args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/click/core.py", line 610, in invoke
    return callback(*args, **kwargs)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tfx/tools/cli/commands/run.py", line 94, in create_run
    handler = handler_factory.create_handler(ctx.flags_dict)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tfx/tools/cli/handler/handler_factory.py", line 93, in create_handler
    return kubeflow_handler.KubeflowHandler(flags_dict)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/tfx/tools/cli/handler/kubeflow_handler.py", line 62, in __init__
    namespace=self.flags_dict[labels.NAMESPACE])
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp/_client.py", line 197, in __init__
    if not self._context_setting['namespace'] and self.get_kfp_healthz(
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp/_client.py", line 411, in get_kfp_healthz
    response = self._healthz_api.get_healthz()
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api/healthz_service_api.py", line 63, in get_healthz
    return self.get_healthz_with_http_info(**kwargs)  # noqa: E501
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api/healthz_service_api.py", line 148, in get_healthz_with_http_info
    collection_formats=collection_formats)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api_client.py", line 369, in call_api
    _preload_content, _request_timeout, _host)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api_client.py", line 185, in __call_api
    _request_timeout=_request_timeout)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/api_client.py", line 393, in request
    headers=headers)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/rest.py", line 234, in GET
    query_params=query_params)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/kfp_server_api/rest.py", line 212, in request
    headers=headers)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/request.py", line 75, in request
    method, url, fields=fields, headers=headers, **urlopen_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/request.py", line 96, in request_encode_url
    return self.urlopen(method, url, **extra_kw)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/poolmanager.py", line 375, in urlopen
    response = conn.urlopen(method, u.request_uri, **kw)
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 826, in urlopen
    **response_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 826, in urlopen
    **response_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 826, in urlopen
    **response_kw
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/connectionpool.py", line 786, in urlopen
    method, url, error=e, _pool=self, _stacktrace=sys.exc_info()[2]
  File "/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/urllib3/util/retry.py", line 592, in increment
    raise MaxRetryError(_pool, url, error or ResponseError(cause))
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='localhost', port=80): Max retries exceeded with url: /apis/v1beta1/healthz (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7ff729e34190>: Failed to establish a new connection: [Errno 111] Connection refused'))

또는 Kubeflow Pipelines 대시보드에서 파이프라인을 실행할 수도 있습니다. 새 실행은 Kubeflow Pipelines 대시보드의 Experiments 아래에 나열됩니다. 실험을 클릭하면 진행 상황을 모니터링하고 실행 실행 중에 생성된 아티팩트를 시각화할 수 있습니다.

Kubeflow Pipelines에서 파이프라인을 실행하는 데 관심이 있는 경우 Cloud AI Platform 파이프라인의 TFX 튜토리얼 에서 자세한 지침을 찾아보세요.

청소

이 단계에서 사용된 모든 GCP 리소스를 정리하기 위해 가이드에 사용한 GCP 프로젝트를 삭제할 수 있습니다.

또는 각 콘솔을 방문하여 개별 리소스를 정리할 수 있습니다.